Driven by the AI boom and antitrust restrictions, new types of deals have taken over Silicon Valley over the past year or so.
I call these transactions “getting steroids” and have the unpleasant flavor of steroid use in sports. If you're a large tech company, you can no longer make a big acquisition for antitrust scrutiny. So instead, they hire top talent, pay handsomely to license technology, and the rest of the business and employees become soldiers alone.
Importantly, Big Tech does not technically acquire these startups, so the transactions are not subject to the same antitrust law. Depending on who you ask, lawyers who have devised this way to avoid the US spirit should win a Nobel Prize for business or be imprisoned.
Remains of prey
Kestrel determines how to divide prey Reuters/MI News/Nur Photo
I'm not here to judge it. I'm focusing on what happens to these startup ruins and why the latest example, Windsurf, is different from the rest of these big Gnarly Acquihires.
Openai was planning to buy Windsurf for $3 billion. The deal fell apart, and instead Google hired Windsurf co-founder Varun Mohan and other senior researchers to pay $2.4 billion to buy the Windsurf license.
For a day or so it looked like the rest of the Windsurf business and staff were thrown into the wind, so to speak. Similar acquisitions of inflections, letters, and scale AI have surprised the ruins of those startups. For example, this week, the Unprofitable Scale AI cut hundreds of jobs after Meta hired CEO Alexandr Wang in a $14 billion deal.
Another result of Windsurf
Windsurf's story has taken a different turn. Another AI startup, Cognition, quickly snapped Windsurf Remnants. why? The answer is data, one of the main components of the success of generator AI.
Related Stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
It certainly requires AI talent. This is the main reason why Google paid a handful of Windsurf leaders and researchers a lot. It also requires infrastructure, such as GPUs, data centers, and huge amounts of electricity. The high-tech giants spend hundreds of millions of millions on it.
The third component, data, is not much talked about. That's because AI companies don't want to pay for their data, so they pretend it's not that important. However, data is important to the development of AI models, and the real reason why Windsurf ruins were purchased so quickly.
IDES in July
Windsurf's main product is the Integrated Development Environment (IDE). The IDE is a coding tool installed on a developer's computer. This is a bit similar to PowerPoint for writing software. They have become the go-to interface between professional programmers and their code.
When Cognition CEO Scott Wu announced his Windsurf contract this week, he explained the main assets he's earning and put Windsurf's IDE at the top of the list. He writes that combining the product with “quick adoption” of AI software engineering assistant Devin will result in “a massive unlock.”
Windsurf CEO Jeff Wang highlighted the same point, saying that the combination of Commition's autonomous coding agent and Windsurf's IDE leads to a “groundbreaking developer experience.”
Fine tuning of unique granular data models
According to Armando Solar-Lezama, a well-known professor of computing at MIT, IDEs are valuable with AI as they provide a large stream of unique granular data on how human developers write, modify, ship and update software code.
“This is why we see so many dill marking activities for some of these startups,” he told me in an interview this week.
Big Tech and AI companies aim to really get the AI models really good at coding. Solar-Lezama said one of the most noticeable ways is to use IDES data at the post-training stage of AI model development.
At a high level, building an AI model involves two main stages.
The first is preoraining for technology companies to clean up all the data on the internet and use it in huge training to help models learn a general understanding of how the world works. Everyone is already stripping the web for this info, so there's no more advantage here in AI races. It's a table stakes.
Then there is a second stage after training. This provides a unique style while AI models models and hones bad behavior. This is where things are different from company to company and you get a conventional and unique approach, Solar-Lezama said.
“All these companies don't have a second internet to mine data after they've exhausted all the data on the internet. They're all looking for alternative data sources,” he said. “One of the major benefits of IDE companies like Cursor and Windsurf is that they can access a wealth of data streams that some AI model providers cannot see directly.”
IDE providers like Windsurf can see every It runs keystrokes and how to run and debug every time the programmer runs the code, and through the IDE installed on the developer's computer.
“This provides unique access to a lot of information about what people are doing with code. It could potentially fall into the prompt box of a simpler coding website.
Large AI companies often provide the underlying model that powers IDE products, allowing them to access some coding data, for example, via API calls. However, this is not detailed information.
“IDEs have a level of granularity that you don't get anywhere else,” says Solar-Lezama.
Sign up for BI's Tech Memo Newsletter here. Please contact me by email abarr@busienssinsider.com.
